lustre/target/tgt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA
  20  *
  21  * GPL HEADER END
  22  */
  23 /*
  24  * Copyright (c) 2011, 2014, Intel Corporation.
  25  */
  26 /*
  27  * lustre/target/tgt_handler.c
  28  *
  29  * Lustre Unified Target request handler code
  30  *
  31  * Author: Brian Behlendorf <behlendorf1@llnl.gov>
  32  * Author: Mikhail Pershin <mike.pershin@intel.com>
  33  */
  34
  35 #define DEBUG_SUBSYSTEM S_CLASS
  36
  37 #include <linux/user_namespace.h>
  38 #ifdef HAVE_UIDGID_HEADER
  39 # include <linux/uidgid.h>
  40 #endif
  41
  42 #include <obd.h>
  43 #include <obd_class.h>
  44 #include <obd_cksum.h>
  45 #include <md_object.h>
  46 #include <lustre_lfsck.h>
  47 #include <lustre_nodemap.h>
  48
  49 #include "tgt_internal.h"
  50
  51 char *tgt_name(struct lu_target *tgt)
  52 {
  53         LASSERT(tgt->lut_obd != NULL);
  54         return tgt->lut_obd->obd_name;
  55 }
  56 EXPORT_SYMBOL(tgt_name);
  57
  58 /*
  59  * Generic code handling requests that have struct mdt_body passed in:
  60  *
  61  *  - extract mdt_body from request and save it in @tsi, if present;
  62  *
  63  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
  64  *  @tsi;
  65  *
  66  *  - if HABEO_CORPUS flag is set for this request type check whether object
  67  *  actually exists on storage (lu_object_exists()).
  68  *
  69  */
  70 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
  71 {
  72         const struct mdt_body   *body;
  73         struct lu_object        *obj;
  74         struct req_capsule      *pill = tsi->tsi_pill;
  75         int                      rc;
  76
  77         ENTRY;
  78
  79         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
  80         if (body == NULL)
  81                 RETURN(-EFAULT);
  82
  83         tsi->tsi_mdt_body = body;
  84
  85         if (!(body->mbo_valid & OBD_MD_FLID))
  86                 RETURN(0);
  87
  88         /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
  89          * in any case in pre-2.5 clients. Fix that here if needed */
  90         if (unlikely(fid_is_zero(&body->mbo_fid1)))
  91                 RETURN(0);
  92
  93         if (!fid_is_sane(&body->mbo_fid1)) {
  94                 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
  95                        PFID(&body->mbo_fid1));
  96                 RETURN(-EINVAL);
  97         }
  98
  99         obj = lu_object_find(tsi->tsi_env,
 100                              &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
 101                              &body->mbo_fid1, NULL);
 102         if (!IS_ERR(obj)) {
 103                 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
 104                         lu_object_put(tsi->tsi_env, obj);
 105                         rc = -ENOENT;
 106                 } else {
 107                         tsi->tsi_corpus = obj;
 108                         rc = 0;
 109                 }
 110         } else {
 111                 rc = PTR_ERR(obj);
 112         }
 113
 114         tsi->tsi_fid = body->mbo_fid1;
 115
 116         RETURN(rc);
 117 }
 118
 119 /**
 120  * Validate oa from client.
 121  * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
 122  * req are valid.
 123  *    a. objects in Single MDT FS  seq = FID_SEQ_OST_MDT0, oi_id != 0
 124  *    b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
 125  *       pack ost_id. Because non-zero oi_seq will make it diffcult to tell
 126  *       whether this is oi_fid or real ostid. So it will check
 127  *       OBD_CONNECT_FID, then convert the ostid to FID for old client.
 128  *    c. Old FID-disable osc will send IDIF.
 129  *    d. new FID-enable osc/osp will send normal FID.
 130  *
 131  * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
 132  * be used for LAST_ID file, and only being accessed inside OST now.
 133  */
 134 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
 135 {
 136         struct ost_id   *oi     = &oa->o_oi;
 137         u64              seq    = ostid_seq(oi);
 138         u64              id     = ostid_id(oi);
 139         int              rc;
 140         ENTRY;
 141
 142         if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
 143                      fid_seq_is_echo(seq))) {
 144                 /* Sigh 2.[123] client still sends echo req with oi_id = 0
 145                  * during create, and we will reset this to 1, since this
 146                  * oi_id is basically useless in the following create process,
 147                  * but oi_id == 0 will make it difficult to tell whether it is
 148                  * real FID or ost_id. */
 149                 oi->oi_fid.f_seq = FID_SEQ_ECHO;
 150                 oi->oi_fid.f_oid = id ?: 1;
 151                 oi->oi_fid.f_ver = 0;
 152         } else {
 153                 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
 154
 155                 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
 156                         GOTO(out, rc = -EPROTO);
 157
 158                 /* Note: this check might be forced in 2.5 or 2.6, i.e.
 159                  * all of the requests are required to setup FLGROUP */
 160                 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
 161                         ostid_set_seq_mdt0(oi);
 162                         oa->o_valid |= OBD_MD_FLGROUP;
 163                         seq = ostid_seq(oi);
 164                 }
 165
 166                 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
 167                                fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
 168                         GOTO(out, rc = -EPROTO);
 169
 170                 rc = ostid_to_fid(&tti->tti_fid1, oi,
 171                                   tsi->tsi_tgt->lut_lsd.lsd_osd_index);
 172                 if (unlikely(rc != 0))
 173                         GOTO(out, rc);
 174
 175                 oi->oi_fid = tti->tti_fid1;
 176         }
 177
 178         RETURN(0);
 179
 180 out:
 181         CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
 182                tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
 183                seq, id, rc);
 184         return rc;
 185 }
 186 EXPORT_SYMBOL(tgt_validate_obdo);
 187
 188 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
 189 {
 190         unsigned                 max_brw;
 191         struct niobuf_remote    *rnb;
 192         struct obd_ioobj        *ioo;
 193         int                      obj_count;
 194
 195         ENTRY;
 196
 197         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
 198         if (ioo == NULL)
 199                 RETURN(-EPROTO);
 200
 201         rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
 202         if (rnb == NULL)
 203                 RETURN(-EPROTO);
 204
 205         max_brw = ioobj_max_brw_get(ioo);
 206         if (unlikely((max_brw & (max_brw - 1)) != 0)) {
 207                 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
 208                        ": rc = %d\n", tgt_name(tsi->tsi_tgt),
 209                        obd_export_nid2str(tsi->tsi_exp), max_brw,
 210                        POSTID(oi), -EPROTO);
 211                 RETURN(-EPROTO);
 212         }
 213         ioo->ioo_oid = *oi;
 214
 215         obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
 216                                         RCL_CLIENT) / sizeof(*ioo);
 217         if (obj_count == 0) {
 218                 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
 219                 RETURN(-EPROTO);
 220         } else if (obj_count > 1) {
 221                 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
 222                        obj_count);
 223                 RETURN(-EPROTO);
 224         }
 225
 226         if (ioo->ioo_bufcnt == 0) {
 227                 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
 228                 RETURN(-EPROTO);
 229         }
 230
 231         if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
 232                 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
 233                           "bulk has too many pages (%d)",
 234                           ioo->ioo_bufcnt);
 235                 RETURN(-EPROTO);
 236         }
 237
 238         RETURN(0);
 239 }
 240
 241 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
 242 {
 243         struct ost_body         *body;
 244         struct req_capsule      *pill = tsi->tsi_pill;
 245         struct lu_nodemap       *nodemap;
 246         int                      rc;
 247
 248         ENTRY;
 249
 250         body = req_capsule_client_get(pill, &RMF_OST_BODY);
 251         if (body == NULL)
 252                 RETURN(-EFAULT);
 253
 254         rc = tgt_validate_obdo(tsi, &body->oa);
 255         if (rc)
 256                 RETURN(rc);
 257
 258         nodemap = tsi->tsi_exp->exp_target_data.ted_nodemap;
 259
 260         body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 261                                         NODEMAP_CLIENT_TO_FS,
 262                                         body->oa.o_uid);
 263         body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 264                                         NODEMAP_CLIENT_TO_FS,
 265                                         body->oa.o_gid);
 266
 267         tsi->tsi_ost_body = body;
 268         tsi->tsi_fid = body->oa.o_oi.oi_fid;
 269
 270         if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
 271                 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
 272                 if (rc < 0)
 273                         RETURN(rc);
 274         }
 275
 276         if (!(body->oa.o_valid & OBD_MD_FLID)) {
 277                 if (flags & HABEO_CORPUS) {
 278                         CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
 279                                "but OID/FID is mandatory with HABEO_CORPUS\n",
 280                                tgt_name(tsi->tsi_tgt));
 281                         RETURN(-EPROTO);
 282                 } else {
 283                         RETURN(0);
 284                 }
 285         }
 286
 287         ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
 288
 289         /*
 290          * OST doesn't get object in advance for further use to prevent
 291          * situations with nested object_find which is potential deadlock.
 292          */
 293         tsi->tsi_corpus = NULL;
 294         RETURN(rc);
 295 }
 296
 297 /*
 298  * Do necessary preprocessing according to handler ->th_flags.
 299  */
 300 static int tgt_request_preprocess(struct tgt_session_info *tsi,
 301                                   struct tgt_handler *h,
 302                                   struct ptlrpc_request *req)
 303 {
 304         struct req_capsule      *pill = tsi->tsi_pill;
 305         __u32                    flags = h->th_flags;
 306         int                      rc = 0;
 307
 308         ENTRY;
 309
 310         if (tsi->tsi_preprocessed)
 311                 RETURN(0);
 312
 313         LASSERT(h->th_act != NULL);
 314         LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
 315         LASSERT(current->journal_info == NULL);
 316
 317         LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
 318                      h->th_fmt != NULL));
 319         if (h->th_fmt != NULL) {
 320                 req_capsule_set(pill, h->th_fmt);
 321                 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
 322                         rc = tgt_mdt_body_unpack(tsi, flags);
 323                         if (rc < 0)
 324                                 RETURN(rc);
 325                 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
 326                                                  RCL_CLIENT)) {
 327                         rc = tgt_ost_body_unpack(tsi, flags);
 328                         if (rc < 0)
 329                                 RETURN(rc);
 330                 }
 331         }
 332
 333         if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
 334                 RETURN(-EROFS);
 335
 336         if (flags & HABEO_CLAVIS) {
 337                 struct ldlm_request *dlm_req;
 338
 339                 LASSERT(h->th_fmt != NULL);
 340
 341                 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 342                 if (dlm_req != NULL) {
 343                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
 344                                      LDLM_IBITS &&
 345                                      dlm_req->lock_desc.l_policy_data.\
 346                                      l_inodebits.bits == 0)) {
 347                                 /*
 348                                  * Lock without inodebits makes no sense and
 349                                  * will oops later in ldlm. If client miss to
 350                                  * set such bits, do not trigger ASSERTION.
 351                                  *
 352                                  * For liblustre flock case, it maybe zero.
 353                                  */
 354                                 rc = -EPROTO;
 355                         } else {
 356                                 tsi->tsi_dlm_req = dlm_req;
 357                         }
 358                 } else {
 359                         rc = -EFAULT;
 360                 }
 361         }
 362         tsi->tsi_preprocessed = 1;
 363         RETURN(rc);
 364 }
 365
 366 /*
 367  * Invoke handler for this request opc. Also do necessary preprocessing
 368  * (according to handler ->th_flags), and post-processing (setting of
 369  * ->last_{xid,committed}).
 370  */
 371 static int tgt_handle_request0(struct tgt_session_info *tsi,
 372                                struct tgt_handler *h,
 373                                struct ptlrpc_request *req)
 374 {
 375         int      serious = 0;
 376         int      rc;
 377         __u32    opc = lustre_msg_get_opc(req->rq_reqmsg);
 378
 379         ENTRY;
 380
 381
 382         /* When dealing with sec context requests, no export is associated yet,
 383          * because these requests are sent before *_CONNECT requests.
 384          * A NULL req->rq_export means the normal *_common_slice handlers will
 385          * not be called, because there is no reference to the target.
 386          * So deal with them by hand and jump directly to target_send_reply().
 387          */
 388         switch (opc) {
 389         case SEC_CTX_INIT:
 390         case SEC_CTX_INIT_CONT:
 391         case SEC_CTX_FINI:
 392                 GOTO(out, rc = 0);
 393         }
 394
 395         /*
 396          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
 397          * to put same checks into handlers like mdt_close(), mdt_reint(),
 398          * etc., without talking to mdt authors first. Checking same thing
 399          * there again is useless and returning 0 error without packing reply
 400          * is buggy! Handlers either pack reply or return error.
 401          *
 402          * We return 0 here and do not send any reply in order to emulate
 403          * network failure. Do not send any reply in case any of NET related
 404          * fail_id has occured.
 405          */
 406         if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
 407                 RETURN(0);
 408
 409         rc = tgt_request_preprocess(tsi, h, req);
 410         /* pack reply if reply format is fixed */
 411         if (rc == 0 && h->th_flags & HABEO_REFERO) {
 412                 /* Pack reply */
 413                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
 414                                           RCL_SERVER))
 415                         req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
 416                                              RCL_SERVER,
 417                                              tsi->tsi_mdt_body->mbo_eadatasize);
 418                 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
 419                                           RCL_SERVER))
 420                         req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
 421                                              RCL_SERVER, 0);
 422
 423                 rc = req_capsule_server_pack(tsi->tsi_pill);
 424         }
 425
 426         if (likely(rc == 0)) {
 427                 /*
 428                  * Process request, there can be two types of rc:
 429                  * 1) errors with msg unpack/pack, other failures outside the
 430                  * operation itself. This is counted as serious errors;
 431                  * 2) errors during fs operation, should be placed in rq_status
 432                  * only
 433                  */
 434                 rc = h->th_act(tsi);
 435                 if (!is_serious(rc) &&
 436                     !req->rq_no_reply && req->rq_reply_state == NULL) {
 437                         DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
 438                                   "pack reply and returned 0 error\n",
 439                                   tgt_name(tsi->tsi_tgt), h->th_name);
 440                         LBUG();
 441                 }
 442                 serious = is_serious(rc);
 443                 rc = clear_serious(rc);
 444         } else {
 445                 serious = 1;
 446         }
 447
 448         req->rq_status = rc;
 449
 450         /*
 451          * ELDLM_* codes which > 0 should be in rq_status only as well as
 452          * all non-serious errors.
 453          */
 454         if (rc > 0 || !serious)
 455                 rc = 0;
 456
 457         LASSERT(current->journal_info == NULL);
 458
 459         if (likely(rc == 0 && req->rq_export))
 460                 target_committed_to_req(req);
 461
 462 out:
 463         target_send_reply(req, rc, tsi->tsi_reply_fail_id);
 464         RETURN(0);
 465 }
 466
 467 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
 468                                        struct obd_device *obd, int *process)
 469 {
 470         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 471         case MDS_DISCONNECT:
 472         case OST_DISCONNECT:
 473         case OBD_IDX_READ:
 474                 *process = 1;
 475                 RETURN(0);
 476         case MDS_CLOSE:
 477         case MDS_SYNC: /* used in unmounting */
 478         case OBD_PING:
 479         case MDS_REINT:
 480         case OUT_UPDATE:
 481         case SEQ_QUERY:
 482         case FLD_QUERY:
 483         case FLD_READ:
 484         case LDLM_ENQUEUE:
 485         case OST_CREATE:
 486         case OST_DESTROY:
 487         case OST_PUNCH:
 488         case OST_SETATTR:
 489         case OST_SYNC:
 490         case OST_WRITE:
 491         case MDS_HSM_PROGRESS:
 492         case MDS_HSM_STATE_SET:
 493         case MDS_HSM_REQUEST:
 494                 *process = target_queue_recovery_request(req, obd);
 495                 RETURN(0);
 496
 497         default:
 498                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
 499                 *process = -EAGAIN;
 500                 RETURN(0);
 501         }
 502 }
 503
 504 /*
 505  * Handle recovery. Return:
 506  *        +1: continue request processing;
 507  *       -ve: abort immediately with the given error code;
 508  *         0: send reply with error code in req->rq_status;
 509  */
 510 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 511 {
 512         ENTRY;
 513
 514         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
 515         case MDS_CONNECT:
 516         case OST_CONNECT:
 517         case MGS_CONNECT:
 518         case SEC_CTX_INIT:
 519         case SEC_CTX_INIT_CONT:
 520         case SEC_CTX_FINI:
 521                 RETURN(+1);
 522         }
 523
 524         if (!req->rq_export->exp_obd->obd_replayable)
 525                 RETURN(+1);
 526
 527         /* sanity check: if the xid matches, the request must be marked as a
 528          * resent or replayed */
 529         if (req_can_reconstruct(req, NULL)) {
 530                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 531                       (MSG_RESENT | MSG_REPLAY))) {
 532                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
 533                                   "saved xid, expected REPLAY or RESENT flag "
 534                                   "(%x)", req->rq_xid,
 535                                   lustre_msg_get_flags(req->rq_reqmsg));
 536                         req->rq_status = -ENOTCONN;
 537                         RETURN(-ENOTCONN);
 538                 }
 539         }
 540         /* else: note the opposite is not always true; a RESENT req after a
 541          * failover will usually not match the last_xid, since it was likely
 542          * never committed. A REPLAYed request will almost never match the
 543          * last xid, however it could for a committed, but still retained,
 544          * open. */
 545
 546         /* Check for aborted recovery... */
 547         if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
 548                 int rc;
 549                 int should_process;
 550
 551                 DEBUG_REQ(D_INFO, req, "Got new replay");
 552                 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
 553                                                  &should_process);
 554                 if (rc != 0 || !should_process)
 555                         RETURN(rc);
 556                 else if (should_process < 0) {
 557                         req->rq_status = should_process;
 558                         rc = ptlrpc_error(req);
 559                         RETURN(rc);
 560                 }
 561         }
 562         RETURN(+1);
 563 }
 564
 565 /* Initial check for request, it is validation mostly */
 566 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
 567 {
 568         struct tgt_handler      *h;
 569         struct tgt_opc_slice    *s;
 570         struct lu_target        *tgt;
 571         __u32                    opc = lustre_msg_get_opc(req->rq_reqmsg);
 572
 573         ENTRY;
 574
 575         tgt = class_exp2tgt(req->rq_export);
 576         if (unlikely(tgt == NULL)) {
 577                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 578                           class_exp2obd(req->rq_export)->obd_name);
 579                 RETURN(ERR_PTR(-EINVAL));
 580         }
 581
 582         for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
 583                 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
 584                         break;
 585
 586         /* opcode was not found in slice */
 587         if (unlikely(s->tos_hs == NULL)) {
 588                 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
 589                        opc);
 590                 RETURN(ERR_PTR(-ENOTSUPP));
 591         }
 592
 593         LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
 594         h = s->tos_hs + (opc - s->tos_opc_start);
 595         if (unlikely(h->th_opc == 0)) {
 596                 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
 597                 RETURN(ERR_PTR(-ENOTSUPP));
 598         }
 599
 600         RETURN(h);
 601 }
 602
 603 int tgt_request_handle(struct ptlrpc_request *req)
 604 {
 605         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 606
 607         struct lustre_msg       *msg = req->rq_reqmsg;
 608         struct tgt_handler      *h;
 609         struct lu_target        *tgt;
 610         int                      request_fail_id = 0;
 611         __u32                    opc = lustre_msg_get_opc(msg);
 612         int                      rc;
 613
 614         ENTRY;
 615
 616         /* Refill the context, to make sure all thread keys are allocated */
 617         lu_env_refill(req->rq_svc_thread->t_env);
 618
 619         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 620         tsi->tsi_pill = &req->rq_pill;
 621         tsi->tsi_env = req->rq_svc_thread->t_env;
 622
 623         /* if request has export then get handlers slice from corresponding
 624          * target, otherwise that should be connect operation */
 625         if (opc == MDS_CONNECT || opc == OST_CONNECT ||
 626             opc == MGS_CONNECT) {
 627                 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
 628                 rc = target_handle_connect(req);
 629                 if (rc != 0) {
 630                         rc = ptlrpc_error(req);
 631                         GOTO(out, rc);
 632                 }
 633                 /* recovery-small test 18c asks to drop connect reply */
 634                 if (unlikely(opc == OST_CONNECT &&
 635                              OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
 636                         GOTO(out, rc = 0);
 637         }
 638
 639         if (unlikely(!class_connected_export(req->rq_export))) {
 640                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
 641                     opc == SEC_CTX_FINI) {
 642                         /* sec context initialization has to be handled
 643                          * by hand in tgt_handle_request0() */
 644                         tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
 645                         h = NULL;
 646                         GOTO(handle_recov, rc = 0);
 647                 }
 648                 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
 649                        opc, libcfs_id2str(req->rq_peer));
 650                 req->rq_status = -ENOTCONN;
 651                 rc = ptlrpc_error(req);
 652                 GOTO(out, rc);
 653         }
 654
 655         tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
 656         tsi->tsi_exp = req->rq_export;
 657         if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
 658                 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
 659         else
 660                 tsi->tsi_jobid = NULL;
 661
 662         if (tgt == NULL) {
 663                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 664                           class_exp2obd(req->rq_export)->obd_name);
 665                 req->rq_status = -EINVAL;
 666                 rc = ptlrpc_error(req);
 667                 GOTO(out, rc);
 668         }
 669
 670         /* check request's xid is consistent with export's last_xid */
 671         if (req->rq_export != NULL) {
 672                 __u64 last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
 673                 if (last_xid != 0)
 674                         req->rq_export->exp_last_xid = last_xid;
 675                 if (req->rq_xid == 0 ||
 676                     req->rq_xid <= req->rq_export->exp_last_xid) {
 677                         DEBUG_REQ(D_ERROR, req,
 678                                   "Unexpected xid %llx vs. last_xid %llx\n",
 679                                   req->rq_xid, req->rq_export->exp_last_xid);
 680 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 93, 0)
 681                         LBUG();
 682 #endif
 683                         req->rq_status = -EPROTO;
 684                         rc = ptlrpc_error(req);
 685                         GOTO(out, rc);
 686                 }
 687         }
 688
 689         request_fail_id = tgt->lut_request_fail_id;
 690         tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 691
 692         /* try to release in-memory reply data */
 693         if (tgt_is_multimodrpcs_client(req->rq_export)) {
 694                 tgt_handle_received_xid(req->rq_export,
 695                                 lustre_msg_get_last_xid(req->rq_reqmsg));
 696                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
 697                       (MSG_RESENT | MSG_REPLAY)))
 698                         tgt_handle_tag(req->rq_export,
 699                                        lustre_msg_get_tag(req->rq_reqmsg));
 700         }
 701
 702         h = tgt_handler_find_check(req);
 703         if (IS_ERR(h)) {
 704                 req->rq_status = PTR_ERR(h);
 705                 rc = ptlrpc_error(req);
 706                 GOTO(out, rc);
 707         }
 708
 709         LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
 710                  h->th_opc, opc);
 711
 712         if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
 713                 GOTO(out, rc = 0);
 714
 715         rc = lustre_msg_check_version(msg, h->th_version);
 716         if (unlikely(rc)) {
 717                 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
 718                           " %08x, expecting %08x\n", tgt_name(tgt),
 719                           lustre_msg_get_version(msg), h->th_version);
 720                 req->rq_status = -EINVAL;
 721                 rc = ptlrpc_error(req);
 722                 GOTO(out, rc);
 723         }
 724
 725 handle_recov:
 726         rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
 727         if (likely(rc == 1)) {
 728                 rc = tgt_handle_request0(tsi, h, req);
 729                 if (rc)
 730                         GOTO(out, rc);
 731         }
 732         EXIT;
 733 out:
 734         req_capsule_fini(tsi->tsi_pill);
 735         if (tsi->tsi_corpus != NULL) {
 736                 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
 737                 tsi->tsi_corpus = NULL;
 738         }
 739         return rc;
 740 }
 741 EXPORT_SYMBOL(tgt_request_handle);
 742
 743 /** Assign high priority operations to the request if needed. */
 744 int tgt_hpreq_handler(struct ptlrpc_request *req)
 745 {
 746         struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
 747         struct tgt_handler      *h;
 748         int                      rc;
 749
 750         ENTRY;
 751
 752         if (req->rq_export == NULL)
 753                 RETURN(0);
 754
 755         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
 756         tsi->tsi_pill = &req->rq_pill;
 757         tsi->tsi_env = req->rq_svc_thread->t_env;
 758         tsi->tsi_tgt = class_exp2tgt(req->rq_export);
 759         tsi->tsi_exp = req->rq_export;
 760
 761         h = tgt_handler_find_check(req);
 762         if (IS_ERR(h)) {
 763                 rc = PTR_ERR(h);
 764                 RETURN(rc);
 765         }
 766
 767         rc = tgt_request_preprocess(tsi, h, req);
 768         if (unlikely(rc != 0))
 769                 RETURN(rc);
 770
 771         if (h->th_hp != NULL)
 772                 h->th_hp(tsi);
 773         RETURN(0);
 774 }
 775 EXPORT_SYMBOL(tgt_hpreq_handler);
 776
 777 void tgt_counter_incr(struct obd_export *exp, int opcode)
 778 {
 779         lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
 780         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
 781                 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
 782 }
 783 EXPORT_SYMBOL(tgt_counter_incr);
 784
 785 /*
 786  * Unified target generic handlers.
 787  */
 788
 789 /*
 790  * Security functions
 791  */
 792 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 793 {
 794         reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 795                                       OBD_CONNECT_RMT_CLIENT_FORCE);
 796 }
 797
 798 static int tgt_init_sec_level(struct ptlrpc_request *req)
 799 {
 800         struct lu_target        *tgt = class_exp2tgt(req->rq_export);
 801         char                    *client;
 802         struct obd_connect_data *data, *reply;
 803         int                      rc = 0;
 804         bool                     remote;
 805         ENTRY;
 806
 807         data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
 808         reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
 809         if (data == NULL || reply == NULL)
 810                 RETURN(-EFAULT);
 811
 812         /* connection from MDT is always trusted */
 813         if (req->rq_auth_usr_mdt) {
 814                 tgt_init_sec_none(reply);
 815                 RETURN(0);
 816         }
 817
 818         if (unlikely(tgt == NULL)) {
 819                 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
 820                           class_exp2obd(req->rq_export)->obd_name);
 821                 RETURN(-EINVAL);
 822         }
 823
 824         client = libcfs_nid2str(req->rq_peer.nid);
 825         /* no GSS support case */
 826         if (!req->rq_auth_gss) {
 827                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 828                         CWARN("client %s -> target %s does not use GSS, "
 829                               "can not run under security level %d.\n",
 830                               client, tgt_name(tgt), tgt->lut_sec_level);
 831                         RETURN(-EACCES);
 832                 } else {
 833                         tgt_init_sec_none(reply);
 834                         RETURN(0);
 835                 }
 836         }
 837
 838         /* old version case */
 839         if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
 840                      !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
 841                      !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
 842                 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
 843                         CWARN("client %s -> target %s uses old version, "
 844                               "can not run under security level %d.\n",
 845                               client, tgt_name(tgt), tgt->lut_sec_level);
 846                         RETURN(-EACCES);
 847                 } else {
 848                         CWARN("client %s -> target %s uses old version, "
 849                               "run under security level %d.\n",
 850                               client, tgt_name(tgt), tgt->lut_sec_level);
 851                         tgt_init_sec_none(reply);
 852                         RETURN(0);
 853                 }
 854         }
 855
 856         remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
 857         if (remote) {
 858                 if (!req->rq_auth_remote)
 859                         CDEBUG(D_SEC, "client (local realm) %s -> target %s "
 860                                "asked to be remote.\n", client, tgt_name(tgt));
 861         } else if (req->rq_auth_remote) {
 862                 remote = true;
 863                 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
 864                        "as remote by default.\n", client, tgt_name(tgt));
 865         }
 866
 867         if (remote == 0) {
 868                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
 869                         CDEBUG(D_SEC, "client %s -> target %s: user is not "
 870                                "authenticated!\n", client, tgt_name(tgt));
 871                         RETURN(-EACCES);
 872                 }
 873         }
 874
 875
 876         switch (tgt->lut_sec_level) {
 877         case LUSTRE_SEC_NONE:
 878                 if (remote) {
 879                         CDEBUG(D_SEC,
 880                                "client %s -> target %s is set as remote, "
 881                                "can not run under security level %d.\n",
 882                                client, tgt_name(tgt), tgt->lut_sec_level);
 883                         RETURN(-EACCES);
 884                 }
 885                 tgt_init_sec_none(reply);
 886                 break;
 887         case LUSTRE_SEC_REMOTE:
 888                 if (!remote)
 889                         tgt_init_sec_none(reply);
 890                 break;
 891         case LUSTRE_SEC_ALL:
 892                 if (remote)
 893                         break;
 894                 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
 895                                               OBD_CONNECT_RMT_CLIENT_FORCE);
 896                 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
 897                 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
 898                 break;
 899         default:
 900                 RETURN(-EINVAL);
 901         }
 902
 903         RETURN(rc);
 904 }
 905
 906 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
 907 {
 908         struct lu_target        *tgt = class_exp2tgt(exp);
 909         struct sptlrpc_flavor    flvr;
 910         int                      rc = 0;
 911
 912         LASSERT(tgt);
 913         LASSERT(tgt->lut_obd);
 914         LASSERT(tgt->lut_slice);
 915
 916         /* always allow ECHO client */
 917         if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
 918                             LUSTRE_ECHO_NAME) == 0)) {
 919                 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 920                 return 0;
 921         }
 922
 923         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
 924                 read_lock(&tgt->lut_sptlrpc_lock);
 925                 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
 926                                              req->rq_sp_from,
 927                                              req->rq_peer.nid,
 928                                              &flvr);
 929                 read_unlock(&tgt->lut_sptlrpc_lock);
 930
 931                 spin_lock(&exp->exp_lock);
 932                 exp->exp_sp_peer = req->rq_sp_from;
 933                 exp->exp_flvr = flvr;
 934
 935                 /* when on mgs, if no restriction is set, or if client
 936                  * is loopback, allow any flavor */
 937                 if ((strcmp(exp->exp_obd->obd_type->typ_name,
 938                            LUSTRE_MGS_NAME) == 0) &&
 939                      (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
 940                       LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid))
 941                       == LOLND))
 942                         exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
 943
 944                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
 945                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
 946                         CERROR("%s: unauthorized rpc flavor %x from %s, "
 947                                "expect %x\n", tgt_name(tgt),
 948                                req->rq_flvr.sf_rpc,
 949                                libcfs_nid2str(req->rq_peer.nid),
 950                                exp->exp_flvr.sf_rpc);
 951                         rc = -EACCES;
 952                 }
 953                 spin_unlock(&exp->exp_lock);
 954         } else {
 955                 if (exp->exp_sp_peer != req->rq_sp_from) {
 956                         CERROR("%s: RPC source %s doesn't match %s\n",
 957                                tgt_name(tgt),
 958                                sptlrpc_part2name(req->rq_sp_from),
 959                                sptlrpc_part2name(exp->exp_sp_peer));
 960                         rc = -EACCES;
 961                 } else {
 962                         rc = sptlrpc_target_export_check(exp, req);
 963                 }
 964         }
 965
 966         return rc;
 967 }
 968
 969 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
 970 {
 971         struct sptlrpc_rule_set  tmp_rset;
 972         int                      rc;
 973
 974         if (unlikely(tgt == NULL)) {
 975                 CERROR("No target passed");
 976                 return -EINVAL;
 977         }
 978
 979         sptlrpc_rule_set_init(&tmp_rset);
 980         rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
 981         if (rc) {
 982                 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
 983                        tgt_name(tgt), rc);
 984                 return rc;
 985         }
 986
 987         sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
 988
 989         write_lock(&tgt->lut_sptlrpc_lock);
 990         sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
 991         tgt->lut_sptlrpc_rset = tmp_rset;
 992         write_unlock(&tgt->lut_sptlrpc_lock);
 993
 994         return 0;
 995 }
 996 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
 997
 998 int tgt_connect(struct tgt_session_info *tsi)
 999 {
1000         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1001         struct obd_connect_data *reply;
1002         int                      rc;
1003
1004         ENTRY;
1005
1006         rc = tgt_init_sec_level(req);
1007         if (rc != 0)
1008                 GOTO(out, rc);
1009
1010         /* XXX: better to call this check right after getting new export but
1011          * before last_rcvd slot allocation to avoid server load upon insecure
1012          * connects. This is to be fixed after unifiyng all targets.
1013          */
1014         rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
1015         if (rc)
1016                 GOTO(out, rc);
1017
1018         /* To avoid exposing partially initialized connection flags, changes up
1019          * to this point have been staged in reply->ocd_connect_flags. Now that
1020          * connection handling has completed successfully, atomically update
1021          * the connect flags in the shared export data structure. LU-1623 */
1022         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
1023         spin_lock(&tsi->tsi_exp->exp_lock);
1024         *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
1025         tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
1026         spin_unlock(&tsi->tsi_exp->exp_lock);
1027
1028         RETURN(0);
1029 out:
1030         obd_disconnect(class_export_get(tsi->tsi_exp));
1031         return rc;
1032 }
1033 EXPORT_SYMBOL(tgt_connect);
1034
1035 int tgt_disconnect(struct tgt_session_info *tsi)
1036 {
1037         int rc;
1038
1039         ENTRY;
1040
1041         rc = target_handle_disconnect(tgt_ses_req(tsi));
1042         if (rc)
1043                 RETURN(err_serious(rc));
1044
1045         RETURN(rc);
1046 }
1047 EXPORT_SYMBOL(tgt_disconnect);
1048
1049 /*
1050  * Unified target OBD handlers
1051  */
1052 int tgt_obd_ping(struct tgt_session_info *tsi)
1053 {
1054         int rc;
1055
1056         ENTRY;
1057
1058         rc = target_handle_ping(tgt_ses_req(tsi));
1059         if (rc)
1060                 RETURN(err_serious(rc));
1061
1062         RETURN(rc);
1063 }
1064 EXPORT_SYMBOL(tgt_obd_ping);
1065
1066 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1067 {
1068         return err_serious(-EOPNOTSUPP);
1069 }
1070
1071 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1072 {
1073         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1074         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1075         struct obd_export       *exp = req->rq_export;
1076         struct ptlrpc_bulk_desc *desc;
1077         struct l_wait_info      *lwi = &tti->tti_u.rdpg.tti_wait_info;
1078         int                      tmpcount;
1079         int                      tmpsize;
1080         int                      i;
1081         int                      rc;
1082
1083         ENTRY;
1084
1085         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1086                                     PTLRPC_BULK_PUT_SOURCE |
1087                                         PTLRPC_BULK_BUF_KIOV,
1088                                     MDS_BULK_PORTAL,
1089                                     &ptlrpc_bulk_kiov_pin_ops);
1090         if (desc == NULL)
1091                 RETURN(-ENOMEM);
1092
1093         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1094                 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1095                  * which is rdpg->rp_count */
1096                 nob = rdpg->rp_count;
1097
1098         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1099              i++, tmpcount -= tmpsize) {
1100                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1101                 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1102                                                  tmpsize);
1103         }
1104
1105         LASSERT(desc->bd_nob == nob);
1106         rc = target_bulk_io(exp, desc, lwi);
1107         ptlrpc_free_bulk(desc);
1108         RETURN(rc);
1109 }
1110 EXPORT_SYMBOL(tgt_sendpage);
1111
1112 /*
1113  * OBD_IDX_READ handler
1114  */
1115 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1116 {
1117         struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
1118         struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1119         struct idx_info         *req_ii, *rep_ii;
1120         int                      rc, i;
1121
1122         ENTRY;
1123
1124         memset(rdpg, 0, sizeof(*rdpg));
1125         req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1126
1127         /* extract idx_info buffer from request & reply */
1128         req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1129         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1130                 RETURN(err_serious(-EPROTO));
1131
1132         rc = req_capsule_server_pack(tsi->tsi_pill);
1133         if (rc)
1134                 RETURN(err_serious(rc));
1135
1136         rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1137         if (rep_ii == NULL)
1138                 RETURN(err_serious(-EFAULT));
1139         rep_ii->ii_magic = IDX_INFO_MAGIC;
1140
1141         /* extract hash to start with */
1142         rdpg->rp_hash = req_ii->ii_hash_start;
1143
1144         /* extract requested attributes */
1145         rdpg->rp_attrs = req_ii->ii_attrs;
1146
1147         /* check that fid packed in request is valid and supported */
1148         if (!fid_is_sane(&req_ii->ii_fid))
1149                 RETURN(-EINVAL);
1150         rep_ii->ii_fid = req_ii->ii_fid;
1151
1152         /* copy flags */
1153         rep_ii->ii_flags = req_ii->ii_flags;
1154
1155         /* compute number of pages to allocate, ii_count is the number of 4KB
1156          * containers */
1157         if (req_ii->ii_count <= 0)
1158                 GOTO(out, rc = -EFAULT);
1159         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1160                                exp_max_brw_size(tsi->tsi_exp));
1161         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1162
1163         /* allocate pages to store the containers */
1164         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1165         if (rdpg->rp_pages == NULL)
1166                 GOTO(out, rc = -ENOMEM);
1167         for (i = 0; i < rdpg->rp_npages; i++) {
1168                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1169                 if (rdpg->rp_pages[i] == NULL)
1170                         GOTO(out, rc = -ENOMEM);
1171         }
1172
1173         /* populate pages with key/record pairs */
1174         rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1175         if (rc < 0)
1176                 GOTO(out, rc);
1177
1178         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1179                  "asked %d > %d\n", rc, rdpg->rp_count);
1180
1181         /* send pages to client */
1182         rc = tgt_sendpage(tsi, rdpg, rc);
1183         if (rc)
1184                 GOTO(out, rc);
1185         EXIT;
1186 out:
1187         if (rdpg->rp_pages) {
1188                 for (i = 0; i < rdpg->rp_npages; i++)
1189                         if (rdpg->rp_pages[i])
1190                                 __free_page(rdpg->rp_pages[i]);
1191                 OBD_FREE(rdpg->rp_pages,
1192                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1193         }
1194         return rc;
1195 }
1196
1197 struct tgt_handler tgt_obd_handlers[] = {
1198 TGT_OBD_HDL    (0,      OBD_PING,               tgt_obd_ping),
1199 TGT_OBD_HDL_VAR(0,      OBD_LOG_CANCEL,         tgt_obd_log_cancel),
1200 TGT_OBD_HDL    (0,      OBD_IDX_READ,           tgt_obd_idx_read)
1201 };
1202 EXPORT_SYMBOL(tgt_obd_handlers);
1203
1204 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1205              struct dt_object *obj, __u64 start, __u64 end)
1206 {
1207         int rc = 0;
1208
1209         ENTRY;
1210
1211         /* if no objid is specified, it means "sync whole filesystem" */
1212         if (obj == NULL) {
1213                 rc = dt_sync(env, tgt->lut_bottom);
1214         } else if (dt_version_get(env, obj) >
1215                    tgt->lut_obd->obd_last_committed) {
1216                 rc = dt_object_sync(env, obj, start, end);
1217         }
1218
1219         RETURN(rc);
1220 }
1221 EXPORT_SYMBOL(tgt_sync);
1222 /*
1223  * Unified target DLM handlers.
1224  */
1225
1226 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1227  * (if requested) */
1228 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1229                             void *data, int flag)
1230 {
1231         struct lu_env            env;
1232         struct lu_target        *tgt;
1233         struct dt_object        *obj;
1234         struct lu_fid            fid;
1235         int                      rc = 0;
1236
1237         ENTRY;
1238
1239         tgt = class_exp2tgt(lock->l_export);
1240
1241         if (unlikely(tgt == NULL)) {
1242                 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1243                        class_exp2obd(lock->l_export)->obd_name);
1244                 RETURN(-EINVAL);
1245         }
1246
1247         if (flag == LDLM_CB_CANCELING &&
1248             (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1249             (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1250              (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1251               lock->l_flags & LDLM_FL_CBPENDING))) {
1252                 __u64 start = 0;
1253                 __u64 end = OBD_OBJECT_EOF;
1254
1255                 rc = lu_env_init(&env, LCT_DT_THREAD);
1256                 if (unlikely(rc != 0))
1257                         RETURN(rc);
1258
1259                 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1260                                    tgt->lut_lsd.lsd_osd_index);
1261                 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1262                 if (IS_ERR(obj))
1263                         GOTO(err_env, rc = PTR_ERR(obj));
1264
1265                 if (!dt_object_exists(obj))
1266                         GOTO(err_put, rc = -ENOENT);
1267
1268                 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1269                         start = lock->l_policy_data.l_extent.start;
1270                         end = lock->l_policy_data.l_extent.end;
1271                 }
1272
1273                 rc = tgt_sync(&env, tgt, obj, start, end);
1274                 if (rc < 0) {
1275                         CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1276                                "cancel: rc = %d\n",
1277                                tgt_name(tgt), PFID(&fid),
1278                                lock->l_policy_data.l_extent.start,
1279                                lock->l_policy_data.l_extent.end, rc);
1280                 }
1281 err_put:
1282                 lu_object_put(&env, &obj->do_lu);
1283 err_env:
1284                 lu_env_fini(&env);
1285         }
1286
1287         rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1288         RETURN(rc);
1289 }
1290
1291 static struct ldlm_callback_suite tgt_dlm_cbs = {
1292         .lcs_completion = ldlm_server_completion_ast,
1293         .lcs_blocking   = tgt_blocking_ast,
1294         .lcs_glimpse    = ldlm_server_glimpse_ast
1295 };
1296
1297 int tgt_enqueue(struct tgt_session_info *tsi)
1298 {
1299         struct ptlrpc_request *req = tgt_ses_req(tsi);
1300         int rc;
1301
1302         ENTRY;
1303         /*
1304          * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1305          * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1306          */
1307         LASSERT(tsi->tsi_dlm_req != NULL);
1308         rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1309                                   tsi->tsi_dlm_req, &tgt_dlm_cbs);
1310         if (rc)
1311                 RETURN(err_serious(rc));
1312
1313         switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1314         case LUT_FAIL_MDT:
1315                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1316                 break;
1317         case LUT_FAIL_OST:
1318                 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1319                 break;
1320         case LUT_FAIL_MGT:
1321                 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1322                 break;
1323         default:
1324                 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1325                 break;
1326         }
1327         RETURN(req->rq_status);
1328 }
1329 EXPORT_SYMBOL(tgt_enqueue);
1330
1331 int tgt_convert(struct tgt_session_info *tsi)
1332 {
1333         struct ptlrpc_request *req = tgt_ses_req(tsi);
1334         int rc;
1335
1336         ENTRY;
1337         LASSERT(tsi->tsi_dlm_req);
1338         rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1339         if (rc)
1340                 RETURN(err_serious(rc));
1341
1342         RETURN(req->rq_status);
1343 }
1344
1345 int tgt_bl_callback(struct tgt_session_info *tsi)
1346 {
1347         return err_serious(-EOPNOTSUPP);
1348 }
1349
1350 int tgt_cp_callback(struct tgt_session_info *tsi)
1351 {
1352         return err_serious(-EOPNOTSUPP);
1353 }
1354
1355 /* generic LDLM target handler */
1356 struct tgt_handler tgt_dlm_handlers[] = {
1357 TGT_DLM_HDL    (HABEO_CLAVIS,   LDLM_ENQUEUE,           tgt_enqueue),
1358 TGT_DLM_HDL_VAR(HABEO_CLAVIS,   LDLM_CONVERT,           tgt_convert),
1359 TGT_DLM_HDL_VAR(0,              LDLM_BL_CALLBACK,       tgt_bl_callback),
1360 TGT_DLM_HDL_VAR(0,              LDLM_CP_CALLBACK,       tgt_cp_callback)
1361 };
1362 EXPORT_SYMBOL(tgt_dlm_handlers);
1363
1364 /*
1365  * Unified target LLOG handlers.
1366  */
1367 int tgt_llog_open(struct tgt_session_info *tsi)
1368 {
1369         int rc;
1370
1371         ENTRY;
1372
1373         rc = llog_origin_handle_open(tgt_ses_req(tsi));
1374
1375         RETURN(rc);
1376 }
1377 EXPORT_SYMBOL(tgt_llog_open);
1378
1379 int tgt_llog_close(struct tgt_session_info *tsi)
1380 {
1381         int rc;
1382
1383         ENTRY;
1384
1385         rc = llog_origin_handle_close(tgt_ses_req(tsi));
1386
1387         RETURN(rc);
1388 }
1389 EXPORT_SYMBOL(tgt_llog_close);
1390
1391
1392 int tgt_llog_destroy(struct tgt_session_info *tsi)
1393 {
1394         int rc;
1395
1396         ENTRY;
1397
1398         rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1399
1400         RETURN(rc);
1401 }
1402
1403 int tgt_llog_read_header(struct tgt_session_info *tsi)
1404 {
1405         int rc;
1406
1407         ENTRY;
1408
1409         rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1410
1411         RETURN(rc);
1412 }
1413 EXPORT_SYMBOL(tgt_llog_read_header);
1414
1415 int tgt_llog_next_block(struct tgt_session_info *tsi)
1416 {
1417         int rc;
1418
1419         ENTRY;
1420
1421         rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1422
1423         RETURN(rc);
1424 }
1425 EXPORT_SYMBOL(tgt_llog_next_block);
1426
1427 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1428 {
1429         int rc;
1430
1431         ENTRY;
1432
1433         rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1434
1435         RETURN(rc);
1436 }
1437 EXPORT_SYMBOL(tgt_llog_prev_block);
1438
1439 /* generic llog target handler */
1440 struct tgt_handler tgt_llog_handlers[] = {
1441 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_CREATE,      tgt_llog_open),
1442 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
1443 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1444 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
1445 TGT_LLOG_HDL    (0,     LLOG_ORIGIN_HANDLE_DESTROY,     tgt_llog_destroy),
1446 TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
1447 };
1448 EXPORT_SYMBOL(tgt_llog_handlers);
1449
1450 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1451                            struct dt_device *key,
1452                            struct lfsck_request *lr,
1453                            struct thandle *th) = NULL;
1454
1455 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1456                                                 struct dt_device *,
1457                                                 struct lfsck_request *,
1458                                                 struct thandle *))
1459 {
1460         tgt_lfsck_in_notify = notify;
1461 }
1462 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1463
1464 static int (*tgt_lfsck_query)(const struct lu_env *env,
1465                               struct dt_device *key,
1466                               struct lfsck_request *lr) = NULL;
1467
1468 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1469                                            struct dt_device *,
1470                                            struct lfsck_request *))
1471 {
1472         tgt_lfsck_query = query;
1473 }
1474 EXPORT_SYMBOL(tgt_register_lfsck_query);
1475
1476 /* LFSCK request handlers */
1477 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1478 {
1479         const struct lu_env     *env = tsi->tsi_env;
1480         struct dt_device        *key = tsi->tsi_tgt->lut_bottom;
1481         struct lfsck_request    *lr;
1482         int                      rc;
1483         ENTRY;
1484
1485         lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1486         if (lr == NULL)
1487                 RETURN(-EPROTO);
1488
1489         rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1490
1491         RETURN(rc);
1492 }
1493
1494 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1495 {
1496         struct lfsck_request    *request;
1497         struct lfsck_reply      *reply;
1498         int                      rc;
1499         ENTRY;
1500
1501         request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1502         if (request == NULL)
1503                 RETURN(-EPROTO);
1504
1505         reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1506         if (reply == NULL)
1507                 RETURN(-ENOMEM);
1508
1509         rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1510         reply->lr_status = rc;
1511
1512         RETURN(rc < 0 ? rc : 0);
1513 }
1514
1515 struct tgt_handler tgt_lfsck_handlers[] = {
1516 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_NOTIFY,   tgt_handle_lfsck_notify),
1517 TGT_LFSCK_HDL(HABEO_REFERO,     LFSCK_QUERY,    tgt_handle_lfsck_query),
1518 };
1519 EXPORT_SYMBOL(tgt_lfsck_handlers);
1520
1521 /*
1522  * initialize per-thread page pool (bug 5137).
1523  */
1524 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1525 {
1526         struct tgt_thread_big_cache *tbc;
1527
1528         ENTRY;
1529
1530         LASSERT(thread != NULL);
1531         LASSERT(thread->t_data == NULL);
1532
1533         OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1534         if (tbc == NULL)
1535                 RETURN(-ENOMEM);
1536         thread->t_data = tbc;
1537         RETURN(0);
1538 }
1539 EXPORT_SYMBOL(tgt_io_thread_init);
1540
1541 /*
1542  * free per-thread pool created by tgt_thread_init().
1543  */
1544 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1545 {
1546         struct tgt_thread_big_cache *tbc;
1547
1548         ENTRY;
1549
1550         LASSERT(thread != NULL);
1551
1552         /*
1553          * be prepared to handle partially-initialized pools (because this is
1554          * called from ost_io_thread_init() for cleanup.
1555          */
1556         tbc = thread->t_data;
1557         if (tbc != NULL) {
1558                 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1559                 thread->t_data = NULL;
1560         }
1561         EXIT;
1562 }
1563 EXPORT_SYMBOL(tgt_io_thread_done);
1564 /**
1565  * Helper function for getting server side [start, start+count] DLM lock
1566  * if asked by client.
1567  */
1568 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1569                     __u64 start, __u64 end, struct lustre_handle *lh,
1570                     int mode, __u64 *flags)
1571 {
1572         ldlm_policy_data_t       policy;
1573         int                      rc;
1574
1575         ENTRY;
1576
1577         LASSERT(lh != NULL);
1578         LASSERT(ns != NULL);
1579         LASSERT(!lustre_handle_is_used(lh));
1580
1581         policy.l_extent.gid = 0;
1582         policy.l_extent.start = start & PAGE_MASK;
1583
1584         /*
1585          * If ->o_blocks is EOF it means "lock till the end of the file".
1586          * Otherwise, it's size of an extent or hole being punched (in bytes).
1587          */
1588         if (end == OBD_OBJECT_EOF || end < start)
1589                 policy.l_extent.end = OBD_OBJECT_EOF;
1590         else
1591                 policy.l_extent.end = end | ~PAGE_MASK;
1592
1593         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1594                                     flags, ldlm_blocking_ast,
1595                                     ldlm_completion_ast, ldlm_glimpse_ast,
1596                                     NULL, 0, LVB_T_NONE, NULL, lh);
1597         RETURN(rc == ELDLM_OK ? 0 : -EIO);
1598 }
1599 EXPORT_SYMBOL(tgt_extent_lock);
1600
1601 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1602 {
1603         LASSERT(lustre_handle_is_used(lh));
1604         ldlm_lock_decref(lh, mode);
1605 }
1606 EXPORT_SYMBOL(tgt_extent_unlock);
1607
1608 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1609                  struct obd_ioobj *obj, struct niobuf_remote *nb,
1610                  struct lustre_handle *lh, int mode)
1611 {
1612         __u64                    flags = 0;
1613         int                      nrbufs = obj->ioo_bufcnt;
1614         int                      i;
1615
1616         ENTRY;
1617
1618         LASSERT(mode == LCK_PR || mode == LCK_PW);
1619         LASSERT(!lustre_handle_is_used(lh));
1620
1621         if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1622                 RETURN(0);
1623
1624         for (i = 1; i < nrbufs; i++)
1625                 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1626                         RETURN(-EFAULT);
1627
1628         RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1629                                nb[nrbufs - 1].rnb_offset +
1630                                nb[nrbufs - 1].rnb_len - 1,
1631                                lh, mode, &flags));
1632 }
1633
1634 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1635                     struct lustre_handle *lh, int mode)
1636 {
1637         ENTRY;
1638
1639         LASSERT(mode == LCK_PR || mode == LCK_PW);
1640         LASSERT((obj->ioo_bufcnt > 0 &&
1641                  (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1642                 lustre_handle_is_used(lh));
1643
1644         if (lustre_handle_is_used(lh))
1645                 tgt_extent_unlock(lh, mode);
1646         EXIT;
1647 }
1648
1649 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1650                                struct ptlrpc_bulk_desc *desc, int opc,
1651                                cksum_type_t cksum_type)
1652 {
1653         struct cfs_crypto_hash_desc     *hdesc;
1654         unsigned int                    bufsize;
1655         int                             i, err;
1656         unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
1657         __u32                           cksum;
1658
1659         LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
1660
1661         hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1662         if (IS_ERR(hdesc)) {
1663                 CERROR("%s: unable to initialize checksum hash %s\n",
1664                        tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1665                 return PTR_ERR(hdesc);
1666         }
1667
1668         CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1669         for (i = 0; i < desc->bd_iov_count; i++) {
1670                 /* corrupt the data before we compute the checksum, to
1671                  * simulate a client->OST data error */
1672                 if (i == 0 && opc == OST_WRITE &&
1673                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1674                         int off = BD_GET_KIOV(desc, i).kiov_offset &
1675                                 ~PAGE_MASK;
1676                         int len = BD_GET_KIOV(desc, i).kiov_len;
1677                         struct page *np = tgt_page_to_corrupt;
1678                         char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1679
1680                         if (np) {
1681                                 char *ptr2 = kmap(np) + off;
1682
1683                                 memcpy(ptr2, ptr, len);
1684                                 memcpy(ptr2, "bad3", min(4, len));
1685                                 kunmap(np);
1686                                 BD_GET_KIOV(desc, i).kiov_page = np;
1687                         } else {
1688                                 CERROR("%s: can't alloc page for corruption\n",
1689                                        tgt_name(tgt));
1690                         }
1691                 }
1692                 cfs_crypto_hash_update_page(hdesc,
1693                                   BD_GET_KIOV(desc, i).kiov_page,
1694                                   BD_GET_KIOV(desc, i).kiov_offset &
1695                                         ~PAGE_MASK,
1696                                   BD_GET_KIOV(desc, i).kiov_len);
1697
1698                  /* corrupt the data after we compute the checksum, to
1699                  * simulate an OST->client data error */
1700                 if (i == 0 && opc == OST_READ &&
1701                     OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1702                         int off = BD_GET_KIOV(desc, i).kiov_offset
1703                           & ~PAGE_MASK;
1704                         int len = BD_GET_KIOV(desc, i).kiov_len;
1705                         struct page *np = tgt_page_to_corrupt;
1706                         char *ptr =
1707                           kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1708
1709                         if (np) {
1710                                 char *ptr2 = kmap(np) + off;
1711
1712                                 memcpy(ptr2, ptr, len);
1713                                 memcpy(ptr2, "bad4", min(4, len));
1714                                 kunmap(np);
1715                                 BD_GET_KIOV(desc, i).kiov_page = np;
1716                         } else {
1717                                 CERROR("%s: can't alloc page for corruption\n",
1718                                        tgt_name(tgt));
1719                         }
1720                 }
1721         }
1722
1723         bufsize = sizeof(cksum);
1724         err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1725
1726         return cksum;
1727 }
1728
1729 int tgt_brw_read(struct tgt_session_info *tsi)
1730 {
1731         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1732         struct ptlrpc_bulk_desc *desc = NULL;
1733         struct obd_export       *exp = tsi->tsi_exp;
1734         struct niobuf_remote    *remote_nb;
1735         struct niobuf_local     *local_nb;
1736         struct obd_ioobj        *ioo;
1737         struct ost_body         *body, *repbody;
1738         struct l_wait_info       lwi;
1739         struct lustre_handle     lockh = { 0 };
1740         int                      npages, nob = 0, rc, i, no_reply = 0;
1741         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1742
1743         ENTRY;
1744
1745         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1746                 CERROR("%s: deny read request from %s to portal %u\n",
1747                        tgt_name(tsi->tsi_tgt),
1748                        obd_export_nid2str(req->rq_export),
1749                        ptlrpc_req2svc(req)->srv_req_portal);
1750                 RETURN(-EPROTO);
1751         }
1752
1753         req->rq_bulk_read = 1;
1754
1755         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1756                 RETURN(-EIO);
1757
1758         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1759                          cfs_fail_val : (obd_timeout + 1) / 4);
1760
1761         /* Check if there is eviction in progress, and if so, wait for it to
1762          * finish */
1763         if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1764                 /* We do not care how long it takes */
1765                 lwi = LWI_INTR(NULL, NULL);
1766                 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1767                          !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1768                          &lwi);
1769         }
1770
1771         /* There must be big cache in current thread to process this request
1772          * if it is NULL then something went wrong and it wasn't allocated,
1773          * report -ENOMEM in that case */
1774         if (tbc == NULL)
1775                 RETURN(-ENOMEM);
1776
1777         body = tsi->tsi_ost_body;
1778         LASSERT(body != NULL);
1779
1780         ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1781         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1782
1783         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1784         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1785
1786         local_nb = tbc->local;
1787
1788         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1789                           remote_nb, &lockh, LCK_PR);
1790         if (rc != 0)
1791                 RETURN(rc);
1792
1793         /*
1794          * If getting the lock took more time than
1795          * client was willing to wait, drop it. b=11330
1796          */
1797         if (cfs_time_current_sec() > req->rq_deadline ||
1798             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1799                 no_reply = 1;
1800                 CERROR("Dropping timed-out read from %s because locking"
1801                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
1802                        libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1803                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1804                        req->rq_deadline - req->rq_arrival_time.tv_sec);
1805                 GOTO(out_lock, rc = -ETIMEDOUT);
1806         }
1807
1808         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1809         repbody->oa = body->oa;
1810
1811         npages = PTLRPC_MAX_BRW_PAGES;
1812         rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1813                         ioo, remote_nb, &npages, local_nb);
1814         if (rc != 0)
1815                 GOTO(out_lock, rc);
1816
1817         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1818                                     PTLRPC_BULK_PUT_SOURCE |
1819                                         PTLRPC_BULK_BUF_KIOV,
1820                                     OST_BULK_PORTAL,
1821                                     &ptlrpc_bulk_kiov_nopin_ops);
1822         if (desc == NULL)
1823                 GOTO(out_commitrw, rc = -ENOMEM);
1824
1825         nob = 0;
1826         for (i = 0; i < npages; i++) {
1827                 int page_rc = local_nb[i].lnb_rc;
1828
1829                 if (page_rc < 0) {
1830                         rc = page_rc;
1831                         break;
1832                 }
1833
1834                 nob += page_rc;
1835                 if (page_rc != 0) { /* some data! */
1836                         LASSERT(local_nb[i].lnb_page != NULL);
1837                         desc->bd_frag_ops->add_kiov_frag
1838                           (desc, local_nb[i].lnb_page,
1839                            local_nb[i].lnb_page_offset,
1840                            page_rc);
1841                 }
1842
1843                 if (page_rc != local_nb[i].lnb_len) { /* short read */
1844                         /* All subsequent pages should be 0 */
1845                         while (++i < npages)
1846                                 LASSERT(local_nb[i].lnb_rc == 0);
1847                         break;
1848                 }
1849         }
1850
1851         if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1852                 cksum_type_t cksum_type =
1853                         cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1854                                           body->oa.o_flags : 0);
1855                 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1856                 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1857                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1858                                                         OST_READ, cksum_type);
1859                 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1860                        repbody->oa.o_cksum);
1861         } else {
1862                 repbody->oa.o_valid = 0;
1863         }
1864         /* We're finishing using body->oa as an input variable */
1865
1866         /* Check if client was evicted while we were doing i/o before touching
1867          * network */
1868         if (likely(rc == 0 &&
1869                    !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1870                 rc = target_bulk_io(exp, desc, &lwi);
1871                 no_reply = rc != 0;
1872         }
1873
1874 out_commitrw:
1875         /* Must commit after prep above in all cases */
1876         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1877                           remote_nb, npages, local_nb, rc);
1878         if (rc == 0)
1879                 tgt_drop_id(exp, &repbody->oa);
1880 out_lock:
1881         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1882
1883         if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1884                 ptlrpc_free_bulk(desc);
1885
1886         LASSERT(rc <= 0);
1887         if (rc == 0) {
1888                 rc = nob;
1889                 ptlrpc_lprocfs_brw(req, nob);
1890         } else if (no_reply) {
1891                 req->rq_no_reply = 1;
1892                 /* reply out callback would free */
1893                 ptlrpc_req_drop_rs(req);
1894                 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1895                               "client will retry: rc %d\n",
1896                               exp->exp_obd->obd_name,
1897                               obd_uuid2str(&exp->exp_client_uuid),
1898                               obd_export_nid2str(exp), rc);
1899         }
1900         /* send a bulk after reply to simulate a network delay or reordering
1901          * by a router */
1902         if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1903                 wait_queue_head_t        waitq;
1904                 struct l_wait_info       lwi1;
1905
1906                 CDEBUG(D_INFO, "reorder BULK\n");
1907                 init_waitqueue_head(&waitq);
1908
1909                 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1910                 l_wait_event(waitq, 0, &lwi1);
1911                 target_bulk_io(exp, desc, &lwi);
1912                 ptlrpc_free_bulk(desc);
1913         }
1914
1915         RETURN(rc);
1916 }
1917 EXPORT_SYMBOL(tgt_brw_read);
1918
1919 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1920                               struct ptlrpc_bulk_desc *desc,
1921                               struct niobuf_local *local_nb, int npages,
1922                               u32 client_cksum, u32 server_cksum,
1923                               bool mmap)
1924 {
1925         struct obd_export *exp = req->rq_export;
1926         struct ost_body *body;
1927         char *router = "";
1928         char *via = "";
1929
1930         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1931         LASSERT(body != NULL);
1932
1933         if (req->rq_peer.nid != desc->bd_sender) {
1934                 via = " via ";
1935                 router = libcfs_nid2str(desc->bd_sender);
1936         }
1937
1938         if (mmap) {
1939                 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1940                              client_cksum, server_cksum);
1941                 return;
1942         }
1943
1944         LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1945                            DFID" object "DOSTID" extent ["LPU64"-"LPU64
1946                            "]: client csum %x, server csum %x\n",
1947                            exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1948                            via, router,
1949                            body->oa.o_valid & OBD_MD_FLFID ?
1950                            body->oa.o_parent_seq : (__u64)0,
1951                            body->oa.o_valid & OBD_MD_FLFID ?
1952                            body->oa.o_parent_oid : 0,
1953                            body->oa.o_valid & OBD_MD_FLFID ?
1954                            body->oa.o_parent_ver : 0,
1955                            POSTID(&body->oa.o_oi),
1956                            local_nb[0].lnb_file_offset,
1957                            local_nb[npages-1].lnb_file_offset +
1958                            local_nb[npages - 1].lnb_len - 1,
1959                            client_cksum, server_cksum);
1960 }
1961
1962 int tgt_brw_write(struct tgt_session_info *tsi)
1963 {
1964         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1965         struct ptlrpc_bulk_desc *desc = NULL;
1966         struct obd_export       *exp = req->rq_export;
1967         struct niobuf_remote    *remote_nb;
1968         struct niobuf_local     *local_nb;
1969         struct obd_ioobj        *ioo;
1970         struct ost_body         *body, *repbody;
1971         struct l_wait_info       lwi;
1972         struct lustre_handle     lockh = {0};
1973         __u32                   *rcs;
1974         int                      objcount, niocount, npages;
1975         int                      rc, i, j;
1976         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
1977         bool                     no_reply = false, mmap;
1978         struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1979
1980         ENTRY;
1981
1982         if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1983                 CERROR("%s: deny write request from %s to portal %u\n",
1984                        tgt_name(tsi->tsi_tgt),
1985                        obd_export_nid2str(req->rq_export),
1986                        ptlrpc_req2svc(req)->srv_req_portal);
1987                 RETURN(err_serious(-EPROTO));
1988         }
1989
1990         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1991                 RETURN(err_serious(-ENOSPC));
1992         if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1993                 RETURN(err_serious(-EROFS));
1994
1995         req->rq_bulk_write = 1;
1996
1997         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1998                 RETURN(err_serious(-EIO));
1999         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
2000                 RETURN(err_serious(-EFAULT));
2001
2002         /* pause before transaction has been started */
2003         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2004                          cfs_fail_val : (obd_timeout + 1) / 4);
2005
2006         /* There must be big cache in current thread to process this request
2007          * if it is NULL then something went wrong and it wasn't allocated,
2008          * report -ENOMEM in that case */
2009         if (tbc == NULL)
2010                 RETURN(-ENOMEM);
2011
2012         body = tsi->tsi_ost_body;
2013         LASSERT(body != NULL);
2014
2015         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2016         LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2017
2018         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2019                                         RCL_CLIENT) / sizeof(*ioo);
2020
2021         for (niocount = i = 0; i < objcount; i++)
2022                 niocount += ioo[i].ioo_bufcnt;
2023
2024         remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2025         LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2026         if (niocount != req_capsule_get_size(&req->rq_pill,
2027                                              &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2028                         sizeof(*remote_nb))
2029                 RETURN(err_serious(-EPROTO));
2030
2031         if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2032             (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
2033                 memory_pressure_set();
2034
2035         req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2036                              niocount * sizeof(*rcs));
2037         rc = req_capsule_server_pack(&req->rq_pill);
2038         if (rc != 0)
2039                 GOTO(out, rc = err_serious(rc));
2040
2041         CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2042         rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2043
2044         local_nb = tbc->local;
2045
2046         rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
2047                           remote_nb, &lockh, LCK_PW);
2048         if (rc != 0)
2049                 GOTO(out, rc);
2050
2051         /*
2052          * If getting the lock took more time than
2053          * client was willing to wait, drop it. b=11330
2054          */
2055         if (cfs_time_current_sec() > req->rq_deadline ||
2056             OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2057                 no_reply = true;
2058                 CERROR("%s: Dropping timed-out write from %s because locking "
2059                        "object "DOSTID" took %ld seconds (limit was %ld).\n",
2060                        tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2061                        POSTID(&ioo->ioo_oid),
2062                        cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2063                        req->rq_deadline - req->rq_arrival_time.tv_sec);
2064                 GOTO(out_lock, rc = -ETIMEDOUT);
2065         }
2066
2067         /* Because we already sync grant info with client when reconnect,
2068          * grant info will be cleared for resent req, then fed_grant and
2069          * total_grant will not be modified in following preprw_write */
2070         if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2071                 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2072                 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2073         }
2074
2075         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2076         if (repbody == NULL)
2077                 GOTO(out_lock, rc = -ENOMEM);
2078         repbody->oa = body->oa;
2079
2080         npages = PTLRPC_MAX_BRW_PAGES;
2081         rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2082                         objcount, ioo, remote_nb, &npages, local_nb);
2083         if (rc < 0)
2084                 GOTO(out_lock, rc);
2085
2086         desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2087                                     PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2088                                     OST_BULK_PORTAL,
2089                                     &ptlrpc_bulk_kiov_nopin_ops);
2090         if (desc == NULL)
2091                 GOTO(skip_transfer, rc = -ENOMEM);
2092
2093         /* NB Having prepped, we must commit... */
2094         for (i = 0; i < npages; i++)
2095                 desc->bd_frag_ops->add_kiov_frag(desc,
2096                                                  local_nb[i].lnb_page,
2097                                                  local_nb[i].lnb_page_offset,
2098                                                  local_nb[i].lnb_len);
2099
2100         rc = sptlrpc_svc_prep_bulk(req, desc);
2101         if (rc != 0)
2102                 GOTO(skip_transfer, rc);
2103
2104         rc = target_bulk_io(exp, desc, &lwi);
2105         no_reply = rc != 0;
2106
2107 skip_transfer:
2108         if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2109                 static int cksum_counter;
2110
2111                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2112                         cksum_type = cksum_type_unpack(body->oa.o_flags);
2113
2114                 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2115                 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2116                 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2117                 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2118                                                         OST_WRITE, cksum_type);
2119                 cksum_counter++;
2120
2121                 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2122                         mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2123                                 body->oa.o_flags & OBD_FL_MMAP);
2124
2125                         tgt_warn_on_cksum(req, desc, local_nb, npages,
2126                                           body->oa.o_cksum,
2127                                           repbody->oa.o_cksum, mmap);
2128                         cksum_counter = 0;
2129                 } else if ((cksum_counter & (-cksum_counter)) ==
2130                            cksum_counter) {
2131                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2132                                cksum_counter, libcfs_id2str(req->rq_peer),
2133                                repbody->oa.o_cksum);
2134                 }
2135         }
2136
2137         /* Must commit after prep above in all cases */
2138         rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2139                           objcount, ioo, remote_nb, npages, local_nb, rc);
2140         if (rc == -ENOTCONN)
2141                 /* quota acquire process has been given up because
2142                  * either the client has been evicted or the client
2143                  * has timed out the request already */
2144                 no_reply = true;
2145
2146         /*
2147          * Disable sending mtime back to the client. If the client locked the
2148          * whole object, then it has already updated the mtime on its side,
2149          * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2150          */
2151         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2152
2153         if (rc == 0) {
2154                 int nob = 0;
2155
2156                 /* set per-requested niobuf return codes */
2157                 for (i = j = 0; i < niocount; i++) {
2158                         int len = remote_nb[i].rnb_len;
2159
2160                         nob += len;
2161                         rcs[i] = 0;
2162                         do {
2163                                 LASSERT(j < npages);
2164                                 if (local_nb[j].lnb_rc < 0)
2165                                         rcs[i] = local_nb[j].lnb_rc;
2166                                 len -= local_nb[j].lnb_len;
2167                                 j++;
2168                         } while (len > 0);
2169                         LASSERT(len == 0);
2170                 }
2171                 LASSERT(j == npages);
2172                 ptlrpc_lprocfs_brw(req, nob);
2173
2174                 tgt_drop_id(exp, &repbody->oa);
2175         }
2176 out_lock:
2177         tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2178         if (desc)
2179                 ptlrpc_free_bulk(desc);
2180 out:
2181         if (no_reply) {
2182                 req->rq_no_reply = 1;
2183                 /* reply out callback would free */
2184                 ptlrpc_req_drop_rs(req);
2185                 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2186                               "client will retry: rc %d\n",
2187                               exp->exp_obd->obd_name,
2188                               obd_uuid2str(&exp->exp_client_uuid),
2189                               obd_export_nid2str(exp), rc);
2190         }
2191         memory_pressure_clr();
2192         RETURN(rc);
2193 }
2194 EXPORT_SYMBOL(tgt_brw_write);
2195
2196 /* Check if request can be reconstructed from saved reply data
2197  * A copy of the reply data is returned in @trd if the pointer is not NULL
2198  */
2199 bool req_can_reconstruct(struct ptlrpc_request *req,
2200                          struct tg_reply_data *trd)
2201 {
2202         struct tg_export_data *ted = &req->rq_export->exp_target_data;
2203         struct lsd_client_data *lcd = ted->ted_lcd;
2204         bool found;
2205
2206         if (tgt_is_multimodrpcs_client(req->rq_export))
2207                 return tgt_lookup_reply(req, trd);
2208
2209         mutex_lock(&ted->ted_lcd_lock);
2210         found = req->rq_xid == lcd->lcd_last_xid ||
2211                 req->rq_xid == lcd->lcd_last_close_xid;
2212
2213         if (found && trd != NULL) {
2214                 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
2215                         trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
2216                         trd->trd_reply.lrd_transno =
2217                                                 lcd->lcd_last_close_transno;
2218                         trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
2219                 } else {
2220                         trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
2221                         trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
2222                         trd->trd_reply.lrd_result = lcd->lcd_last_result;
2223                         trd->trd_reply.lrd_data = lcd->lcd_last_data;
2224                         trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
2225                         trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
2226                         trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
2227                         trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
2228                 }
2229         }
2230         mutex_unlock(&ted->ted_lcd_lock);
2231
2232         return found;
2233 }
2234 EXPORT_SYMBOL(req_can_reconstruct);
2235